Exercise 0. Although Python(x,y) comes already with a great variety of scientic Python packages, we might have to install additional dependencies:

Transcription

1 Exercise 0 Deadline: None Computer Setup Windows Download Python(x,y) via and install it. Make sure that before installation the installer does not complain about an existing Python version on your computer. Although Python(x,y) comes already with a great variety of scientic Python packages, we might have to install additional dependencies: For instance, we will make use of the vigranumpy package which you can download and install from this website: Now you can test whether your installation was successful: Find the Python(x,y) folder in the Windows start menu and launch the contained Spyder program. In the interactive interpreter on the bottom right, enter import import import vigra numpy sklearn If no output appears, your installation was successful. If the sklearn import failed, try to install the latest version from Ubuntu Either compile everything yourself, use pip, or just execute the following line to install the system libraries we need: sudo apt - get install build - essential python - dev python - numpy python - setuptools python - scipy libatlas - dev libatlas3 - base python - matplotlib python - vigra python - sklearn spyder If you want to start GUI programs such as spyder, always start them from the terminal; this way they inherit all necessary environment variables that have been set in /.bashrc. Mac Install Homebrew ( by running this command: ruby -e "$( curl - fssl https :// raw. github. com / mxcl / homebrew / go )" If you get errors here, follow these instructions: Then, insert the Homebrew directory at the top of your PATH variable by adding the following line at the bottom of your /.bash_profile le: export PATH =/ usr / local / bin : $PATH Now open a new terminal and run the following command: brew doctor 1/6

2 As long as you do not get Your system is ready to brew, try to x the warnings. In the next step, you can install Python using Homebrew via brew install python You can then install most python packages using pip: pip install numpy pip install scikit - learn pip install matplotlib pip install scipy pip install scikit - image pip install spyder Note that it is not possible to install vigranumpy like that. In case you need this dependency for your project, we can assist you installing it. Troubleshooting: If pip install scipy fails, you might have to run brew install gfortran rst. If pip install matplotlib fails, try to do the following brew install freetype brew install libpng brew link -- force freetype and then reinstall scipy. The Scientic Python Ecosystem In this course, we are going to use the Python programming language. There exist many highly usable and well maintained scientic libraries for Python. Packages that will be useful for us are: numpy (numpy.org), provides multi dimensional arrays and fast numeric routines that work with these arrays. scipy (scipy.org), a collection of many scientic algorithms for areas such as optimization, linear algebra, integration, interpolation, FFT, signal and image processing. Makes use of numpy arrays. matplotlib (matplotlib.org), a plotting library which provides a MATLAB like interface. Check out the great gallery with many examples. scikit-learn (scikit-learn.org), a quickly growing collection of machine learning algorithms, such as Support Vector Machines, Decision Trees, Nearest Neighbor Methods and many more. Their website oers good overview documentation and great examples, too. scikit-image (scikit-image.org), a collection of image processing algorithms. vigranumpy (hci.iwr.uni-heidelberg.de/vigra, a C++ library for multidimensional arrays, image processing and machine learning, developed in our group. It exposes most functions to Python via the vigranumpy module. All of these packages can be easily installed on recent Linux distributions, such as Ubuntu. Coming from MATLAB? Read Indexing starts at zero in numpy! IPython For an interactive python shell, start ipython within a shell. It supports auto-completion using the tab key, as well as showing the documentation of functions by appending a? to the function name. 2/6

3 Spyder Spyder ( is a MATLAB-like IDE for scientic Python. You can use the editor on the left or the interactive interpreter on the bottom right, just like in MATLAB. Python For a tutorial introduction to Python, we recommend sections 1 through 6 as well as 7.2 and 9.3. The following sub-sections, which are aimed at advanced users, can be skipped: 2.2, 4.7, 5.1.3, 5.1.4, 5.6, 5.7, 5.8, 6.4. In the following, we give a quick overview of the language in order to get you started quickly. Indentation Python uses indentation to group statements as opposed to curly braces in C-like languages. Always indent using 4 spaces. Also note that braces around statements following if, for, etc. are omitted; instead, a colon : at the end is used. if a == 42: for i in range (5) : print i else : b = 5 Variables Python, as a scripting language, uses dynamic typing. a = 42 # int b1, b2 = 42.0, float (42) # float, cast to float c = MyClass ( a) # instance of MyClass d = None # special ' none ' type e = " Hello " # string f = [1,2,3] # a list of values ( mutable ), f [0]=1 g = (1,2,3) # a tuple of values ( immutable ), g [0]=1 h = {"x": 1, "y": 2} #a dictionary h['x ']=1, h['y ']=2 Functions Functions are declared with def. They take a list of required arguments and optional keyword arguments (with default values), and may return values. def some_func ( arg1, arg2, kwarg1 =True, kwarg2 =42) : return arg1 + arg2 Control Flow if, elif, else: if ( a or b) and c: print "x" elif b: print "y" else : print "z" A for loop: for i in range (0,10) : # range [0,10) print i if i < 2: continue if i >= 7: break Exercise: Fizz-Buzz ( 3/6

4 Write a program that prints the numbers from 1 to 100. But for multiples of three print Fizz instead of the number and for the multiples of ve print Buzz. For numbers which are multiples of both three and ve print FizzBuzz. Lists, tuples and slicing List, tuple or strings can be subscripted using slice notation. s = " Hello " print s [1:] print s [2:4] print s [0: -1:2] print s [::2] Exercise: Can you explain the output? Modules Additional functionality can be accessed by importing modules. If you want to import only parts of a module, use the from module import symbol notation, where the function can be optionally renamed using as: import math from math import pi as circle_number print " cos ( pi ) =% f" % ( math. cos ( circle_number ),) Every python le you write can be imported from you in other les: If le file1.py denes a function function1, you can use it from file2.py in the same directory via from file1 import function1. Files Write import statements always at the beginning of your python le. Always put your main function in if name == " main " at the end of the le. This way, the main code won't be executed if you import this le from another le. Numpy A good tutorial for numpy can be found at Import the numpy module via import numpy Creating and calculating with arrays import numpy # create a 2D array of shape (4,6), data type is unsigned char a = numpy. zeros ((4,6), dtype = numpy. uint8 ) print a. ndim # 2- dimensional print a. shape, a. shape [0], a. shape [1] print a. dtype # numpy. uint8 # uniform random numbers in [0,1) b = numpy. random. random (a. shape ) # basic algebra for arrays c = a+b d = a*b e = a /( b +1) f = numpy. sqrt (b) Array slicing Slicing works similar to the slicing of lists, tuples and strings as introduced above. For multidimensional arrays, a slicing for each dimension is given (comma separated). Please see http: //docs.scipy.org/doc/numpy/reference/arrays.indexing.html for a comprehensive overview. 4/6

5 a [:] = 1 # write a 1 to every entry a [1,2] = 2 # assign 2 to the row 1, col 2 ( indices start at 0!) s = numpy. sum ( a) # sum over all array elements assert s == 7 a [:,0] = 42 # assign 42 to the 0- th column # ':' == > select all rows # '0' == > select the 0- th column a [0,...] = 42 # fill 0- th element of first array dimension ( here : column ) # with 42 's, no matter how many dimensions a has b = a [:,0:2] # make a subarray : select only columns 0 and 1 # ( range [0..2) ) numpy.where To nd the indices in an array where a certain condition matches, use numpy.where. The returned tuple of indices can be passed as an argument to the [] operator of any array: import numpy a = numpy. asarray ([[2,5,4,3,1],[1,2,1,5,7]]) print a. shape print a print numpy. where ( a == 2) print numpy. where ( a == 4) print a[ numpy. where (a == 1) ] Vigra Import the vigranumpy module using import vigra. The documentation can be found at index.html. vigranumpy may be used, among many other functions, to read in images as numpy arrays or write them out to le 1 : import vigra img = vigra. impex. readimage ( './ myimage. png ') img. shape # red channel img = img [...,0] img. shape vigra. impex. writeimage (img, './ redchannel. png ') Note that many functions expect the array to be of a certain data type. If you get errors like Boost.Python.ArgumentError: Python argument types in vigra.filters.gaussiansmoothing(numpy.ndarray, int) did not match C++ signature: gaussiansmoothing(vigra::numpyarray<4u, vigra::multiband<float>, vigra::stridedarraytag> array, boost::python::api::object sigma, vigra::numpyarray<4u, vigra::multiband<float>, vigra::stridedarraytag> out=none, boost::python::api::object sigma_d=0.0, boost::python::api::object step_size=1.0, double window_size=0.0, boost::python::api::object roi=none) gaussiansmoothing(vigra::numpyarray<3u, vigra::multiband<float>, vigra::stridedarraytag> array, boost::python::api::object sigma, vigra::numpyarray<3u, vigra::multiband<float>, vigra::stridedarraytag> out=none, boost::python::api::object sigma_d=0.0, boost::python::api::object step_size=1.0, double window_size=0.0, boost::python::api::object roi=none) you have forgotten to cast to the correct data type (note that it searches above for a vigra::numpyarray<4u, vigra::multiband<float>, vigra::stridedarraytag>, which has pixel type oat). In this case, you can cast to 32-bit oat using array.astype(numpy.float32). Other vigra functions may use data type numpy.uint8 or numpy.uint32. Additionally, check that your array has the correct shape, in this case, the function expects an array with ndim == 3 or ndim == 4. 1 If you did not manage to install vigra on your machine, you might want to check out scikit-image for some of those operations. 5/6

6 Matplotlib The recommended way to import matplotlib is: import matplotlib ; matplotlib. use (" Qt4Agg ") from mpl_toolkits. mplot3d import Axes3D from matplotlib import pyplot as plot See to copy/paste some code. Example: read and display image. import numpy, vigra import matplotlib ; matplotlib. use ( ' Qt4Agg ') from matplotlib import pyplot as plot img = vigra. impex. readimage (" myimage. png ") [...,0] # read in red channel plot. figure () plot. gray () # select grayscale color map plot. imshow ( img ) plot. show () Scikit-learn Scikit-learn is a very powerful machine learning toolbox. Most of the algorithms contained in this package will be reviewed in the Pattern Recognition lecture. A tutorial for scikit-learn including detailed descriptions of the algorithms is available at index.html. Exercises (optional, no pts) 1. Plot sin(x) for x [ 1, 1]. Label the axes and add a title. Then save the gure as png. Implementation hints: Use numpy and matplotlib.pyplot. 2. Write a function computing n! and call the function from your main function for n = Find and output the eigenvalues and eigenvectors of the following matrix: A = (1) Implementation hints: Check out the documentation page of numpy.linalg for eigenvalue computation. 4. Concatenate the eigenvectors as columns in a matrix P, compute P 1, and conrm that P Λ P 1 == A, where Λ is the diagonal matrix of the eigenvalues of A. What does a * p do? Implementation hints: In numpy, nd a function for matrix multiplication and for creating a diagonal matrix. Matrix inversion is implemented in the numpy.linalg package. You might also want to look at numpy.all. 6/6